Method and apparatus for detecting charged state of secondary battery based on neural network calculation

Abstract

A neural network type of apparatus is provided to detect an internal state of a secondary battery implemented in a battery system. The apparatus comprises a detecting unit, producing unit and estimating unit. The detecting unit detects electric signals indicating an operating state of the battery. The producing unit produces, using the electric signals, an input parameter required for estimating the internal state of the battery. The input parameter reflects calibration of a present charged state of the battery which is attributable to at least one of a present degraded state of the battery and a difference in types of the battery. The estimating unit estimates an output parameter indicating the charged state of the battery by applying the input parameter to neural network calculation.

Description

CROSS REFERENCES TO RELATED APPLICATIONS[0001]The present application relates to and incorporates by reference Japanese Patent application Nos. 2005-036442 filed on Feb. 14, 2005, 2005-036437 filed on Feb. 14, 2005, 2005-039614 filed on Feb. 16, 2005, 2005-122011 filed on Apr. 20, 2005, 2005-122004 filed on Apr. 20, 2005, and 2005-151050 filed on May 24, 2005.BACKGROUND OF THE INVENTION[0002]1. Field of the Invention[0003]The present invention relates to a battery system with a neural network type of apparatus for detecting a charged state of a secondary (rechargeable) battery, and in particular, to an improvement in detection of the charged state of such a battery which is for example mounted on vehicles.[0004]2. Description of the Related Art[0005]An on-vehicle battery system is mostly composed of a secondary battery such as lead batteries. In the secondary battery, degrees of degradation give fluctuations to correlations between electric quantities of a battery, such as voltage a...

AI Technical Summary

Benefits of technology

[0019]Thus, the foregoing fundamental configuration adopts the technique of including, into the input parameters for neural network calibration, data of calibration of a present charged state of a battery, the calibration reflecting, as described, for example, the degree of degradation in charge / discharge of a used battery. Including such calibration data into the input parameters enables an output parameter to be calculated (estimated) more accurately than, for example, a situation where data of voltage and current history are simply used as input parameters. By repeating such estimation at intervals, the charged state of a used battery can be checked with high precision, with automatically tracking temporal degradations of the charge / discharge performance of the used battery.

[0023]In this way, by adding, to the input parameter, only an open-circuit voltage of the battery and an internal resistance of the battery, which are detected responsively to a discharge of a predetermined amount of power from a battery which has been fully charged, the precision for the neural network calculation can be done with high precision, while still preventing the number of input parameters from being increased. Accordingly, the size of a neural network calculator can be kept smaller, but the full charge capacity of a used battery can be calculated with precision, even compared to conventional calculators with or without a neural network. And the time for calculation can be kept in a period of time required for practical use. As a result, with no paying attention to over-charge and over-discharge, the capacity range for use can be widened. Compared to the conventional, a battery can be made more compact, while still being enough for covering a necessary discharge capacity range for the battery. This will lead to not merely less space occupation for mounting a battery on vehicles but also a decrease in the vehicle body weight. In consequence, the second object of the present invention can also be attained.

[0025]Selectively using the plurality of memory tables described above makes it possible that coupling coefficients for neural network calculation are selected to have a higher correlation with presently acquired input parameters, increasing accuracy in calculating a charged state of a battery. This accurate calculation can be realized, provided that a memory capacity for this calculation is allowed to increase slightly, thus the size of circuitry for the calculation being prevented from increasing. A rise in the processing time for the calculation is almost never required, because selectively reading the memory tables requires only changing the addresses of the coupling coefficients in a memory. Delay in the calculation will not occur. Hence, though the memory capacity increases a little, the charged state of a battery can be detected with higher precision.

[0027]In this aspect, a large number of pairs of sampled voltage and current data are not necessary for the calculation, resulting in that the circuitry can be avoided from increasing in its size and a calculation load can be reduced, and the precision of neural network calculation is also secured. To be more specific, the input parameters include the function value f(Vo, R) whose variables Vo and R are individually correlated with a degradation and a residual capacity of each battery, respectively. In other words, compared with sole use of the open-circuit voltage Vo or internal resistance R, the function value f(Vo, R), which shows a dischargeable amount of power, has higher correlations with degraded states and charged states of each battery. Thus, the influence of a battery degradation on its residual capacity is well reflected in the input parameters, which leads to the above advantage.

Problems solved by technology

These problems make it difficult to detect, with precision, the SOC and / or SOH of each of secondary batteries which are mass-produced.

However, even when the SOC and / or SOH are detected using the conventional neural network type of detection apparatus, fluctuations and changes in the measurement precision, which are due to degradations in the battery, cannot be prevented sufficiently.

The existence of those various different correlations makes it difficult to absorb the fluctuations and changes in the measurement precision even when calculation is made using the neural network.

However, even when the present values of such physical quantities are taken into account as part of the input parameters, a substantial progress in the degradation of the battery makes it difficult to attain or keep a practically-required higher level in detecting the SOC and / or SOH.

However, such a configuration is not favorable, because the calculator becomes large in its circuit size, a calculation load increases, and much power is consumed.

Images